Can AI help with patient selection for Aduhelm and other new disease-modifying drugs for Alzheimer’s disease?

Alzheimer’s disease (AD) is the most common type of dementia. It’s a progressive disease beginning with mild memory loss and leading to more severe disruptions in memory, loss of the ability to carry on with daily activities, and ultimately to death. In 2020, as many as 5.8 million Americans were living with AD, and this number is projected to nearly triple to 14 million people by 20601. The situation is similar in all other developed countries. The costs to society are enormous: The direct costs of AD treatment are estimated at $305 billion, while the cost of informal care is an estimated $244 billion2. However, because many people with AD are undiagnosed3, estimating the true costs of AD is difficult.
There is no cure for AD, and until recently, approved pharmacological treatments for AD have focused only on reducing symptoms rather than halting the degenerative processes. These treatments include rivastigmine, galantamine, donepezil, memantine, and memantine+donepezil. The effects vary by person and can wane over time4.

What is it about AD that makes drug development challenging?

Describing “typical” AD is not easy. Disease progression and characteristics differ significantly from person to person and can be affected by age, genetics, and other factors. Early clinical symptoms include difficulty remembering recent conversations, names, or events; apathy; and depression. As the disease progresses, impaired communication, disorientation, confusion, poor judgment, behavioral changes, and difficulty speaking, swallowing, and walking can also occur.
For brain pathology, AD causes loss of neurons and their networks, resulting in brain atrophy and loss of brain volume. As part of this process, beta-amyloid protein forms plaques outside neurons (the hallmark of AD), and the protein tau forms twisted strands inside neurons, both of which can cause inflammation. These changes are thought to begin decades before symptoms become apparent.
Therefore, AD is thought to be a continuum consisting of pre-clinical AD, mild cognitive impairment (MCI) due to AD, and dementia due to AD. Each phase is characterized by differences in imaging and other biomarkers.
Source: Alzheimer’s Association Report: 2020 Alzheimer’s disease facts and figures. Alzheimer’s & Dementia 16:391-460.

Why is the right patient selection for Aduhelm and drugs in clinical practice and clinical trials important?

Lacking an effective drug against AD, there is clearly a huge unmet need. The pharmaceutical industry has spent billions over the last couple of decades trying to develop disease-modifying drugs (DMDs) that can, if not stop, at least significantly delay disease progression in AD by targeting the upstream pathobiology.
However, this has been a very difficult task, and several hundreds of AD drug trials have failed in the past two decades. It was therefore a big relief to the field when the first DMD in AD, Biogen’s aducanumab (marketed under the name Aduhelm®) was conditionally approved by the FDA in June 20215. This was despite conflicting results in two of the study arms stemming from early study termination and a narrow study sample. This has also resulted in approval for only MCI due to AD or mild AD.

One reason for trial failure is inappropriate patient selection. In many studies, the drug was administered too late in the disease course, or the sample was too heterogeneous. In other studies, the study duration was too short to prove a drug effect. In some cases, the need to reduce the cost of a lengthy trial could have contributed to shortened timelines.

In AD studies, costs are associated with initial testing for patient selection and the amount of testing needed for safety follow-up and to assess study endpoints. For example, confirmation of amyloid (or changes in amyloid) require positron emission tomography (PET) or cerebrospinal fluid (CSF) tests. Clearly, it is desirable to avoid expensive testing in patients that will not be included in the trial (e.g., patients that turn out to be amyloid-negative), so any technology that can help stratify patients using low-cost tests would be useful. Furthermore, it is critical to exclude patients who are not likely to respond to the treatment being studied, whether due to genetic, disease duration, disease phase, or other factors. Therefore, selecting the patients with the right disease profile is crucial.
of patients are eligible for Aduhelm

based on 540 patients in three memory clinics in the Combinostics database

How can we ensure appropriate patient selection for Aduhelm and other DMDs in development?

Finding the right patient in the most cost-effective way is key to the success of DMDs, be it in clinical trials or in clinical use. It’s challenging for humans to process the broad range of data required to confidently identify patients meeting the complex criteria for a DMD trial or treatment: imaging biomarkers (such as presence of atrophy and amyloid), genetic characteristics, other biomarkers, cognitive test results, and demographic information. Further, determining how likely, and how quickly, a patient’s condition will progress is nearly impossible to determine. This is where AI can help.
Machine learning systems can be trained on large amounts of data from patients that have known diagnoses and outcomes.
For a new patient:

The specific diagnosis (AD vs other forms of dementia) and likely progression can be determined by providing the machine learning model with all of that patient’s available data, against which it can compare with the database of previous patients’ data.

For DMDs:

Such a system could be trained to predict the likelihood of the patient being amyloid-positive.

If amyloid results are available, they can be included in the predictive analytics to determine whether the patient is on a slow or fast track for cognitive decline.

Follow-up measurements provided to the system can help determine the actual rate of decline, at a specificity not achievable by humans.

For clinical trials:

This type of system could help identify patients who have the required characteristics (i.e., amyloid plaque) and are likely to respond to the drug, reducing the “noise” associated with disease heterogeneity.

For treatment with an approved drug:

It can provide evidence for the suitability of that drug for the specific patient.

When multiple DMDs against AD are available on the market, it can be used to decide which drug is likely to work best for each patient.

So then, our answer to the question posed in the title is, quite simply, yes. The use of AI for patient selection is not a pie-in-the-sky idea — the capabilities already exist in our cDSI product, which is part of the cNeuro platform. We’re certain that AI can — and will — help pharma companies select and monitor trial participants during drug development, as well as help clinicians decide whether a patient is eligible for a certain type of treatment.

  1. Matthews, K. A., Xu, W., Gaglioti, A. H., Holt, J. B., Croft, J. B., Mack, D., & McGuire, L. C. (2018). Racial and ethnic estimates of Alzheimer’s disease and related dementias in the United States (2015–2060) in adults aged≥ 65 years. Alzheimer’s & Dementia 15(1):17-24.
  2. Wong, W. (2020). Economic Burden of Alzheimer Disease and Managed Care Considerations. Am J Manag Care 26:S177-S183.
  3. Gauthier, S., Rosa-Neto, P., Morais, J.A., & Webster, C. 2021. World Alzheimer Report 2021: Journey through the diagnosis of dementia. London, England: Alzheimer’s Disease International.
  4. Atri, A. (2019). Current and Future Treatments in Alzheimer’s Disease. Semin Neurol 39(2):227-240.
  5. Food & Drug Administration. (June 7, 2021). FDA Grants Accelerated Approval for Alzheimer’s Drug. Available at: